
An AI feature can help with a defined task without being appropriate for every document or decision. Before using it with a hybrid team, identify the information involved, the people affected and the person accountable for the use. Convenience is not evidence of privacy or compliance.
Start with the data flow
List what enters the tool, where processing happens, who can access inputs and outputs, and how long each is kept. Check whether the provider can reuse prompts or documents for training. Submit only information you are authorised to share; use a less sensitive example when it serves the task. A local deployment can reduce some transfers, but still needs access controls and maintenance.
Assess the controls, not the label
Ask how deletion, permissions, incident handling and changes to vendor terms work in practice. On-device processing, federated learning and differential privacy can limit particular exposures under specific designs; none makes every use anonymous or secure. Review the actual configuration and residual risks, rather than relying on a privacy-friendly product description.
Explain the use and check the applicable rules
Tell people what the system does and how to raise a concern. Transparency alone does not establish compliance with the GDPR, CCPA or other applicable law. The purpose, legal basis, necessity, rights, contracts and transfers may need separate assessment. Involve the organisation's privacy and security specialists before processing sensitive information or using AI in consequential decisions.
Keep accountable human review
Treat generated content as a draft. Check facts against original sources and allow errors to be corrected before they affect a client or colleague. A person approving an output needs the time, authority and information to question it. Neither automation nor a nominal human approval guarantees accuracy, fairness or a better decision.
Retire outdated examples
FLoC is an abandoned browser advertising proposal, not a current workplace privacy solution. Google also announced the retirement of Topics and several other Privacy Sandbox technologies in October 2025; that announcement did not end every Privacy Sandbox API. Evaluate the current documentation of the actual tool you intend to use. No example here certifies a vendor or FLYDESK as compliant.
Sources checked on 2026-10-10: CNIL — Generative AI questions and answers · NIST — AI Risk Management Framework · Google — Privacy Sandbox update (2025) · Google — FLoC archive
Original illustrations and translated captions remain historical material; they do not certify a current product or privacy control.
